AI Model Stability in Industrial IoT Intrusion Detection: Leveraging the Characteristics Stability Index

Abstract

In Industrial Internet of Things (IIoT) environments, the reliability and adaptability of machine learning models are crucial for accurate decision-making. This paper introduces the Characteristic Stability Index (CSI) to monitor and ensure the stability of models in the context of heterogeneous IIoT sensor data. The CSI quantifies the variations in feature importance rankings, enabling the early detection of data drift and shifts. The experimentation results validate the performance of the decision tree algorithm to provide actionable insights, facilitating domain experts’ adaptability and enhancing decision-making while minimizing operational risks and costs in the choice of intrusion detection systems model.